Dengue Is Hacked by Team DatAsset

THE TEAM

Klaas Nelissen

Pieter De Koninck

Michael Reusens

Andreas Withaeckx

Joren Raymenants

THE PROBLEM

Spatio-temporal predictive modelling of Dengue evolution in Sri Lanka

  • Why Sri Lanka?
  • Abrupt increase of Dengue burden in last couple of years
  • Able to build upon data preparation efforts - Thank You!

APPROACH

  1. Visualization - What is going on at the province level?

  2. Time Series Modeling - How many people when and where?

  3. Business Understanding - Can we leverage expert knowledge?

THE VISUALIZATION

TIME SERIES MODELLING PER PROVINCE: EXAMPLE OF VARIATION

TIME SERIES FORECASTING PER PROVINCE

We used:

  • Historic number of Dengue cases per province
  • Climate data
  • Worldbank Country Descriptors
    • Sanitation
    • Development (e.g. slums)
    • Water

-> MAPE of +-36% for out-of-time-validation

Lots of room for improvement!

LR-test for Worldbank var. has p-value of 0.12

TIME SERIES FORECASTING ERROR

The Bigger Picture

IF NOTHING ELSE WORKS, WE CAN STILL BUILD A WALL